Nvidia’s Future: Can It Steer Clear of a Downfall?

The current landscape of AI data centers hinges on two critical factors: Nvidia chips and borrowed capital. It seems almost inevitable that the demand for Nvidia chips would lead to increased borrowing. As the excitement around AI continues to grow, I’ve started to feel uneasy about the vulnerabilities within this booming sector; a closer look at the financial dynamics has done little to ease my concerns.

Nvidia has made significant investments in the AI realm, with over 70 investments in AI companies this year alone, according to PitchBook data. A notable area of focus is neoclouds, like CoreWeave, a publicly traded company heavily in debt and banking on the idea that data centers will keep expanding indefinitely. Companies like CoreWeave have taken on debt to acquire Nvidia chips, using the chips themselves as collateral, effectively transforming $1 of Nvidia investment into $5 worth of chip purchases. While this is beneficial for Nvidia, I’m not so sure it’s a win for everyone else involved.

There’s been extensive discussion about how quickly these chips depreciate and whether that rapid decline makes the loans nonsensical. While I appreciate the enthusiasm behind this analysis, it somewhat misses the larger picture: Nvidia has a vested interest in propping up this industry as most GPU-backed loans are secured by its own chips. This creates a precarious situation; if Nvidia stumbles, the entire sector could face serious repercussions, especially with rising competition for its chips.

Can startups keep pace with chip depreciation, and is it happening faster than they’re letting on? Loans secured by depreciating assets aren’t a new concept. For those deeply entrenched in finance, GPUs are seen as interchangeable commodities, akin to trucks or houses. This allows companies like CoreWeave to bundle chips with AI contracts and other assets to secure loans, typically for purchasing more chips. If they default, lenders can reclaim the collateral, similar to how banks repossess homes.

Lenders often mitigate risks by adjusting interest rates based on perceived risk. For instance, a 10-year mortgage currently sits at around 5.3 percent, while CoreWeave’s first GPU-backed loan in 2023 had an interest rate of 14 percent. Another strategy is to require a high percentage of collateral relative to the loan, known as the loan-to-value (LTV) ratio. If I buy a house for $500,000, I might put down 20 percent, resulting in an LTV of 80 percent.

LTV ratios for GPU loans can vary significantly, influenced by loan duration, confidence in management, and other factors. Some loans have LTVs as low as 50 percent, while others can exceed 100 percent. The majority of these loans are tied to Nvidia chips, which could further entrench Nvidia’s market dominance. If a company needs GPUs, it might secure better financing terms on Nvidia’s chips due to their liquidity.

Not everyone is convinced about the viability of these loans. Keri Findley, CEO of Tacora Capital, expresses skepticism, noting that while she understands the depreciation curve for cars, she’s uncertain about GPUs. She suspects they may lose value rapidly, especially if they need to be leased multiple times before being worth investing in. The secondary market for older chips remains unclear.

Determining the value and lifespan of GPUs is more complex than it is for a house. CoreWeave has indicated that its borrowing capacity is tied to the value of its GPUs, which will decline as their worth diminishes. If GPUs depreciate faster than anticipated, CoreWeave may need to inject more capital into its loans.

Some investors, including notable short-seller Michael Burry, argue that many companies are overly optimistic about the longevity of their GPUs, estimating a $176 billion discrepancy in depreciation between 2026 and 2028. While Burry’s focus isn’t primarily on neoclouds, they are particularly susceptible. Hyperscalers like Google and Amazon can absorb write-downs without significant damage due to their diversified operations, whereas neoclouds lack that cushion. They may face write-downs and complications with their costly loans, potentially needing to provide more collateral when demand for their services wanes.

Trinity Capital is holding onto its loans, betting that even if some AI companies fail, demand for the chips will persist. If a neocloud goes bankrupt due to miscalculating chip depreciation, its customers might continue their operations while banks reclaim the servers, selling them at a loss. This scenario wouldn’t spell disaster for lenders or customers, but it would negatively impact Nvidia by flooding the market with used chips and reducing its customer base. A simultaneous failure of multiple companies would exacerbate the situation.

So, how vulnerable is Nvidia? The surge in AI lending is partly driven by private credit firms eager to deliver returns for their investors while competing with one another. If they misjudge the risks associated with GPU loans, the fallout could extend to banks, potentially causing broader economic chaos.

Earlier, we discussed interest rates as a reflection of risk pricing. However, there’s another perspective: viewing loans as products driven by supply and demand. For lenders not planning to hold loans long-term, risk assessment may take a backseat to making and flipping loans.

Consider a hypothetical neocloud startup, WarSieve, seeking a GPU-backed loan. If I’m unsure about the market for their GPUs, I might set a high interest rate. If WarSieve accepts, I could sell the loan to clients interested in high-yield debt. If a competitor offers a better rate, I’d have to lower mine to stay competitive, focusing more on securing deals than on assessing risk.

Several factors are fueling the AI lending boom. Analysts predict $3 trillion in spending on data centers by 2028, coinciding with private credit managers sitting on large cash reserves but struggling to make deals. This creates an environment ripe for pouring funds into AI.

CoreWeave has been a major player in GPU-backed loans, securing a $2.3 billion loan initially backed by private credit firms. Following that, it took out additional loans totaling billions. Other companies, like Fluidstack and Crusoe, have also secured substantial loans, with even mid-sized GPU loans reaching tens of millions.

Many of these borrowers are startups, emulating CoreWeave’s strategy of rapid growth through debt. Fluidstack, for instance, reported only $65 million in revenue for 2024 but still secured a massive loan. The private credit market has expanded significantly, with GPU-backed loans offering higher interest rates than some junk bonds, making them particularly appealing.

Private credit can also benefit established companies by creating special-purpose vehicles that allow them to take on debt without affecting their credit ratings. This unregulated space has led to the creation of asset-backed securities and derivative financial products, reminiscent of the mortgage derivatives that contributed to the 2008 financial crisis.

While the GPU debt slice is relatively small compared to Big Tech’s bond issuances, it reflects broader trends in tech lending. The tech sector has taken on more debt than during the dot-com bubble, raising concerns about the risks associated with private debt, which is generally riskier than bank debt.

AI companies create a link between private credit and traditional banks, raising the stakes for AI lending beyond just the fortunes of individual firms. For example, CoreWeave has a $2.5 billion revolving credit line with a major bank, highlighting the interconnectedness of these financial systems.

Private debt also impacts banks directly, as they often lend to private credit providers. As of June, banks had lent $300 billion to these firms, and aggressive growth could weaken underwriting standards, increasing credit risk.

“Borrowing by AI companies should be on the radar as a potential threat to the financial system and broader economy,” warns an economist. Unlike the dot-com boom, where exuberance was primarily in equity, the current debt landscape means that if AI falters, the repercussions could be widespread.

AI investments have significantly contributed to GDP growth, with some analysts suggesting the economy might already be in recession without them. However, the current AI debt load is not expected to directly trigger a financial crisis.

The interconnected nature of the AI sector heightens its vulnerability, with Nvidia playing a central role through its investments and chip sales. The issue of depreciation extends beyond just chips; it’s a complex interplay of financial dynamics.

Debt is fundamentally about numbers, while equity often hinges on sentiment. This is why concerns about GPUs losing value faster than anticipated are so pressing. The debate centers on whether older chips become economically unviable after three years, while many companies depreciate them over five or six years. This discrepancy affects earnings and the assumptions underlying GPU-backed loans.

The financial implications are significant. For instance, if I run a company like Live Wire Server Farms and invest in Nvidia Tesla V100s, I could recoup my investment in just a few months, depending on usage. However, operational costs, including power and cooling, add complexity to the equation.

Newer chips are more efficient but also require more power, which is a critical limitation for the industry. The demand for additional power capacity is projected to grow, but new sources are lagging behind, potentially extending the lifespan of older chips.

Data centers require specialized infrastructure, and the challenges of power and construction may lead to older chips remaining in use longer. However, at some point, the operational costs of older GPUs will exceed what I can charge customers, necessitating new investments.

If I need to secure a GPU loan to build out infrastructure for the next generation of chips, I’m relying on my existing GPUs and contracts as collateral. As long as everything runs smoothly, this is manageable. But unexpected risks can disrupt the entire sector, as seen with Bitcoin miners who faced massive losses when the market turned.

A similar scenario could unfold with GPU-backed loans, but the stakes are higher due to the involvement of traditional banks. If multiple neoclouds fail, the market could be flooded with used chips, impacting Nvidia significantly.

The AI lending boom is also driven by private credit firms eager to deliver returns while competing with one another. If they misjudge the risks associated with GPU loans, the fallout could extend to banks, potentially causing broader economic chaos.

The interconnectedness of the AI sector raises the stakes for AI lending beyond just the fortunes of individual firms. If several neoclouds collapse, the market could be flooded with data centers full of chips, leading to significant repercussions for Nvidia and the broader economy.

The potential for systemic financial crises is heightened by the interconnectedness of players, concentration of risk, and regulatory gaps. The AI industry is highly interconnected, with many companies taking out loans on assets with uncertain valuations. This creates a precarious situation, and the consequences of a collapse could ripple through the economy.

The question isn’t just how the music stops, but when it does and what follows.